When cursor agent runs terminal commands in powershell, it has 2 failure modes that happen very often:
- it keyboard interrupts the commands after not very long a lot of the time
- the command STDOUT/STDERR aren’t properly mapped back to the composer
both of these make using the composer very difficult/iffy in agent mode
Thanks for the report!
We’re hoping to improve the inline terminal compatibility soon, but we have noticed the AI can be optimistic about using keyboard interrupts in the terminals!
For now, try adding to your Rules for AI an instruction to not use keyboard interrupts, as this may help avoid this in the short term.
I have done this several times, it’s not helped (though the AI definitely gets more apologetic the more intense I make the rule phrasing when it inevitably interrupts again)